Computer Vision for Augmented Reality in Handheld Devices
- Version
- Download 20
- File Size 625.87 KB
- Download
Computer Vision for Augmented Reality in Handheld Devices
Dheeraj Vaddepally
dheeraj.vaddepally@gmail.com
Abstract— Augmented Reality (AR) on mobile devices has picked up great momentum in a host of industries, providing engaging and interactive experiences via the blending of digital overlays with real-world environments. A key enabling technology behind such applications is computer vision, which facilitates real-time object detection and tracking for AR overlays. Yet, implementing effective computer vision models on low-powered handheld devices raises a host of challenges. This work discusses lightweight object detection and tracking techniques optimized for mobile platforms with the focus on such models as MobileNet and YOLO-tiny that are capable of providing real-time performance with very low computational costs. Another major concern in mobile AR systems is balancing the requirements of graphic rendering for AR overlays with machine learning (ML) computations for object detection. This paper reviews methods of dividing resources among these mutually competing processes, i.e., model pruning, adaptive rendering, and hybrid processing methodologies. Methods of reducing latency and optimizing power efficiency are also presented to improve the user experience on mobile devices. Additionally, privacy and security issues are discussed in the realm of AR with edge computing identified as a way to protect sensitive information. The paper concludes by outlining future directions for AR on handheld devices, such as improvements in computer vision models, the development of AR-specific hardware, and new applications taking advantage of this technology in education, healthcare, and entertainment. Through this, we hope to give a holistic view of challenges and opportunities involved in deploying computer vision for augmented reality on mobile platforms.
Keywords— augmented reality, computer vision, lightweight object detection, mobile AR, real-time tracking, graphic rendering, machine learning, resource-constrained devices, edge computing, privacy.